FedRio: Personalized Federated Social Bot Detection via Cooperative Reinforced Contrastive Adversarial Distillation

Yingguang Yang; Hao Liu; Xin Zhang; Yunhui Liu; Yutong Xia; Qi Wu; Hao Peng; Taoran Liang

cs.AIApr 12, 2026

FedRio: Personalized Federated Social Bot Detection via Cooperative Reinforced Contrastive Adversarial Distillation

Authors: Yingguang Yang, Hao Liu, Xin Zhang, Yunhui Liu, Yutong Xia, Qi Wu, Hao Peng, Taoran Liang, +3 more

Organizations: School of Cyber science at University of Science and Technology of China, Hefei 230026, China · State Key Laboratory for Novel Software Technology, Nanjing University, Nanjing 210023, China · Institute of Data Science at the National University of Singapore, 119077, Singapore · State Key Laboratory of Software Development Environment, Beihang University, Beijing 1000191, China · in Systems Science from Beijing University of Posts and Telecommunications, Beijing 100876, China · National Engineering Laboratory for Big Data Analysis and Applications, Peking University, Beijing, 100871, P.R. China · Department of Computer Science, University of Illinois Chicago, Chicago, IL 60607, USA

Abstract

Social bot detection is critical to the stability and security of online social platforms. However, current state-of-the-art bot detection models are largely developed in isolation, overlooking the benefits of leveraging shared detection patterns across platforms to improve performance and promptly identify emerging bot variants. The heterogeneity of data distributions and model architectures further complicates the design of an effective cross-platform and cross-model detection framework. To address these challenges, we propose FedRio (Personalized Federated Social Bot Detection with Cooperative Reinforced Contrastive Adversarial Distillation framework. We first introduce an adaptive message-passing module as the graph neural network backbone for each client. To facilitate efficient knowledge sharing of global data distributions, we design a federated knowledge extraction mechanism based on generative adversarial networks. Additionally, we employ a multi-stage adversarial contrastive learning strategy to enforce feature space consistency among clients and reduce divergence between local and global models. Finally, we adopt adaptive server-side parameter aggregation and reinforcement learning-based client-side parameter control to better accommodate data heterogeneity in heterogeneous federated settings. Extensive experiments on two real-world social bot detection benchmarks demonstrate that FedRio consistently outperforms state-of-the-art federated learning baselines in detection accuracy, communication efficiency, and feature space consistency, while remaining competitive with published centralized results under substantially stronger privacy constraints.

Figures & tables

s_{t}=[r_{t}^{1},\mathbf{p}_{t}^{1},\ell_{t}^{1},\ldots,r_{t}^{K},\mathbf{p}_{t}^{K},\ell_{t}^{K}]

Algorithm 2 Client Parameter Update Based on Reinforcement Learning

Dataset	Total Nodes	Edges	Bots	Humans
Vendor-19	5,349	11,284	2,553	2,796
TwiBot-20	11,826	85,927	6,589	5,237

TABLE I: Statistics of the Datasets

Panel of Figure 4 — Fig. 2: Visualization of data heterogeneity. The darker color means more training samples with a label available to the client.

Metric	Accuracy				F1
Setting	$\alpha=1$	$\alpha=0.5$	$\alpha=0.1$	$\alpha=0.05$	$\alpha=1$	$\alpha=0.5$	$\alpha=0.1$	$\alpha=0.05$
FedAvg	71.30 $\pm$ 0.60	61.06 $\pm$ 1.52	60.88 $\pm$ 2.85	59.81 $\pm$ 2.48	68.48 $\pm$ 4.08	66.78 $\pm$ 0.15	65.70 $\pm$ 0.65	66.61 $\pm$ 0.01
FedProx	84.37 $\pm$ 0.43	78.25 $\pm$ 1.02	51.86 $\pm$ 0.04	63.27 $\pm$ 2.32	84.99 $\pm$ 0.60	76.70 $\pm$ 0.67	77.16 $\pm$ 0.61	67.07 $\pm$ 0.15
FedDF	86.37 $\pm$ 1.23	80.17 $\pm$ 2.21	63.16 $\pm$ 1.37	67.01 $\pm$ 1.78	-	-	-	-
FedEnsemble	81.12 $\pm$ 2.22	76.70 $\pm$ 1.21	64.51 $\pm$ 2.56	68.05 $\pm$ 1.15	80.87 $\pm$ 8.04	67.77 $\pm$ 1.20	69.45 $\pm$ 3.38	67.35 $\pm$ 0.40
FedDistill	79.68 $\pm$ 0.58	68.77 $\pm$ 1.13	52.88 $\pm$ 0.06	70.25 $\pm$ 0.39	77.20 $\pm$ 0.09	69.16 $\pm$ 1.00	67.08 $\pm$ 0.37	67.57 $\pm$ 0.17

TABLE II: Comparison of the average maximum accuracy of different methods for social bot detection on Dataset Vendor-19 (%).

Metric	Accuracy				F1
Setting	$\alpha=1$	$\alpha=0.5$	$\alpha=0.1$	$\alpha=0.05$	$\alpha=1$	$\alpha=0.5$	$\alpha=0.1$	$\alpha=0.05$
FedAvg	54.04 $\pm$ 0.50	55.41 $\pm$ 1.35	51.37 $\pm$ 0.77	52.46 $\pm$ 0.02	53.37 $\pm$ 0.01	55.06 $\pm$ 2.96	50.40 $\pm$ 0.00	50.41 $\pm$ 0.01
FedProx	74.34 $\pm$ 0.06	73.32 $\pm$ 0.25	51.86 $\pm$ 0.04	52.30 $\pm$ 0.63	78.46 $\pm$ 0.17	77.33 $\pm$ 0.45	50.08 $\pm$ 0.00	51.02 $\pm$ 0.25
FedDF	72.12 $\pm$ 1.96	71.25 $\pm$ 1.03	55.23 $\pm$ 1.32	53.35 $\pm$ 1.41	-	-	-	-
FedEnsemble	55.98 $\pm$ 2.55	54.15 $\pm$ 0.04	54.21 $\pm$ 0.04	54.15 $\pm$ 0.04	68.63 $\pm$ 16.88	45.73 $\pm$ 5.23	47.38 $\pm$ 5.53	38.99 $\pm$ 1.74
FedDistill	64.11 $\pm$ 0.29	63.34 $\pm$ 0.56	50.00 $\pm$ 0.00	54.30 $\pm$ 0.05	66.04 $\pm$ 1.59	63.99 $\pm$ 0.27	50.08 $\pm$ 0.00	52.50 $\pm$ 0.37

TABLE III: Comparison of the average maximum accuracy of different methods for social bot detection on Dataset TwiBot-20 (%).

Panel of Figure 7 — Fig. 3: Loss dynamics during training under different data consistency parameters and loss functions.

Panel of Figure 8 — Fig. 4: Hyperparameter Sensitivity ( $\gamma,\tau,\mu$ ) of FedACK on Vendor-19 under diferent data heterogeneity settings ( $\alpha$ ).

Panel of Figure 9 — Fig. 5: Learning Curve of (a-b) Vendor-19 and (c-d) TwiBot-20 in 100 communication rounds in different $\alpha$ settings.

Dataset	Vendor-19		TwiBot-20
Setting	$\alpha=1\ (85)$	$\alpha=0.5\ (75)$	$\alpha=1\ (75)$	$\alpha=0.5\ (70)$
FedAvg	unreached	unreached	unreached	unreached
FedProx	unreached	50.2 $\pm$ 1.3	13.3 $\pm$ 2.8	44.4 $\pm$ 8.4
FedDF	unreached	unreached	unreached	unreached
FedDistill	unreached	60.3 $\pm$ 12.6	unreached	unreached
FedGen	25.2 $\pm$ 3,7	7.5 $\pm$ 1.2	10.6 $\pm$ 0.9	16.6 $\pm$ 2.2

TABLE IV: The round number to reach the target accuracy on Vendor-19

(80\%,70\%)

and TwiBot-20

(70\%,65\%)

Panel of Figure 11 — Fig. 6: Decision boundaries and feature space of two randomly selected clients from FedRio trained on Vendor-19. The x-axis and y-axis represent the values of the 2-dimensional features output by $\varepsilon$ described in Sec V-D .

Dataset	Vendor-19
Setting	$\alpha=1$	$\alpha=0.5$	$\alpha=0.1$	$\alpha=0.05$
FedRio -NA&NR&NC	88.58 $\pm$ 1.91	87.05 $\pm$ 2.03	76.04 $\pm$ 3.40	75.27 $\pm$ 2.50
FedRio -NA&NR	93.28 $\pm$ 0.94	87.28 $\pm$ 01.32	77.92 $\pm$ 1.78	70.41 $\pm$ 1.73
FedRio -NA	94.19 $\pm$ 1.23	89.37 $\pm$ 0.92	78.02 $\pm$ 2.16	77.57 $\pm$ 1.93
FedRio -NC	89.38 $\pm$ 1.45	88.31 $\pm$ 1.63	76.31 $\pm$ 1.27	71.12 $\pm$ 1.72
FedRio -NR	92.37 $\pm$ 1.03	90.07 $\pm$ 0.79	73.89 $\pm$ 2.19	70.27 $\pm$ 2.53

TABLE V: Comparison of the average maximum accuracy (%) of different methods with/without AGA (NA), with/without RL (NR), and with/without the adaptive message-passing module (NC). Gain is the disparity between FedRio and other baselines.

Method	Setting	Acc.(%)	F1(%)	Data Access
Centralized methods (full TwiBot-20, 229K nodes, complete graph)
SGBot †	Central.	79.58	83.51	Full graph + tweets
BotRGCN †	Central.	84.42	86.90	Full graph + metadata
RGT †	Central.	84.70	87.19	Full graph + metadata
SimpleHGN †	Central.	84.65	87.13	Full heterog. graph
LMBot †	Central.	85.63	87.61	Full graph $\to$ LM distill.

TABLE VI: Contextual comparison between centralized social bot detectors (published results on TwiBot-20 with full data access) and FedRio (federated setting with local subgraph access only). Centralized results are cited from the original papers. “

\dagger

” indicates results cited from LMBot. [ 36 ] .

Yingguang Yang received the Ph.D. degree in the School of Cyber Science at University of Science and Technology of China. His research interests include machine learning, data mining, and social bot detection.

Table 14

Hao Liu is currently a master’s student in the School of Cyber Science and Technology at the University of Science and Technology of China. His research interests include fairness, federated learning and graph neural networks.

Table 15

Xin Zhang is currently a master’s student in the School of Cyber Science and Technology at the University of Science and Technology of China. Her research interests include fairness, federated learning and privacy policy security under the guidance of LLM.

Table 16

Yunhui Liu is working toward the Ph.D. degree with the Software Institute, Nanjing University, China. His research interests include graph machine learning and self-supervised learning.

Table 17

Yutong Xia is currently a PhD student at the Institute of Data Science, National University of Singapore. She has published several papers in refereed conferences, such as NeurIPS, IJCAI, AAAI, and SIGSPATIAL. Her research interests mainly lie in spatio-temporal data mining and graph neural networks.

Table 18

Qi Wu is currently a master’s student in the School of Cyber Science and Technology at University of Science and Technology of China. His research interests include multi-view learning, social bot detection and graph neural networks.

Table 19

Hao Peng is currently a Professor at the School of Cyber Science and Technology in Beihang University. His current research interests include machine learning, deep learning, and reinforcement learning. He is the Associate Editor of the International Journal of Machine Learning and Cybernetics (IJMLC).

Table 20

Taoran Liang received the M.Sc. degree in Systems Science from Beijing University of Posts and Telecommunications, China. His research interests include reinforcement learning for large language models (LLMs) and graph neural networks (GNNs).

Table 21

Bin Chong received the Ph.D. degree in physical chemistry from the Peking University. He is currently an associate professor in the Peking University. His research interest includes machine learning, molecular modelling and digital transformation.

Table 22

Tieke He received the B.E. and Ph.D. degrees in software engineering from the Software Institute, Nanjing University, Jiangsu, China. He is currently an Associate Professor with the Software Institute, Nanjing University. His research interests lie in intelligent software engineering, knowledge graph, and question answering.

Table 23

Philip S. Yu is a Distinguished Professor and the Wexler Chair in Information Technology at the Department of Computer Science, University of Illinois Chicago. He is a Fellow of the ACM and IEEE. Dr. Yu has published more than 1,200 referred conference and journal papers cited more than 190,000 times with an H-index of 195. He has applied for more than 300 patents. Dr. Yu was the Editor-in-Chief of ACM TKDD (2011-2017) and IEEE TKDE (2001-2004).

Table 24

Explore similar work

RABot: Reinforcement-Guided Graph Augmentation for Imbalanced and Noisy Social Bot Detection
Feb 25, 2026Longlong Zhang, Xi Wang, Haotong Du +3Data Augmentation Class-Imbalanced Learning
School of Artificial Intelligence, Optics and Electronics (iOPEN), Northwestern Polytechnical University · Shenzhen Research Institute, Northwestern Polytechnical University · School of Automation Science and Engineering, Xi’an Jiaotong University +1
Byzantine-Robust Federated Fire Detection with a Rotating Coordinator
Sep 9, 2026Georgia Argyrou, Aymen Bahroun, Hedi Fendri +1Federated Learning Aggregation Byzantine-Robust Federated Learning
Department of Computer Science, Aalto University, Espoo, Finland · Kudelski Labs
Social bot detection in the age of ChatGPT: Challenges and opportunities
Oct 1, 2026Emilio FerraraSocial Media Analysis AI-Generated Text Detection

Feb 25, 2026J/K move · Enter open · S save

RABot: Reinforcement-Guided Graph Augmentation for Imbalanced and Noisy Social Bot Detection

Longlong Zhang, Xi Wang, Haotong Du, Yangyi Xu+2

School of Artificial Intelligence, Optics and Electronics (iOPEN), Northwestern Polytechnical University · Shenzhen Research Institute, Northwestern Polytechnical University · School of Automation Science and Engineering, Xi’an Jiaotong University+1

Data Augmentation Class-Imbalanced Learning Graph Neural Networks Graph Structure Learning

Open paper AskPDF arXiv

Social bot detection is pivotal for safeguarding the integrity of online information ecosystems. Although recent graph neural network (GNN) solutions achieve strong results, they remain hindered by two practical challenges: (i) severe class imbalance arising from the high cost of generating bots, and (ii) topological noise introduced by bots that skillfully mimic human behavior and forge deceptive links. We propose the Reinforcement-guided graph Augmentation social Bot detector (RABot), a multi-granularity graph-augmentation framework that addresses both issues in a unified manner. RABot employs a neighborhood-aware oversampling strategy that linearly interpolates minority-class embeddings within local subgraphs, thereby stabilizing the decision boundary under low-resource regimes. Concurrently, a reinforcement-learning-driven edge-filtering module combines similarity-based edge features with adaptive threshold optimization to excise spurious interactions during message passing, yielding a cleaner topology. Extensive experiments on three real-world benchmarks and four GNN backbones demonstrate that RABot consistently surpasses state-of-the-art baselines. In addition, since its augmentation and filtering modules are orthogonal to the underlying architecture, RABot can be seamlessly integrated into existing GNN pipelines to boost performance with minimal overhead.

Abstract

Explore similar work

RABot: Reinforcement-Guided Graph Augmentation for Imbalanced and Noisy Social Bot Detection

Byzantine-Robust Federated Fire Detection with a Rotating Coordinator

Social bot detection in the age of ChatGPT: Challenges and opportunities